Anyone have encountered case like this?

Have you considered either using the AIUtil suggested element that is at the bottom of your inspect window which would allow you to interact with the rendered screen, or considered TextUtil for OCR support?

There have been times that certain objects in mobile are not easily addressable.    This is one of the many reasons our team shifted to the AI engine using nearly all AIUtil statements in our test code.

for the mobile app that I tested, the AI is not able to detect all objects 100% accurately. It's often in the first iteration the AI could detect the object but in the next iteration, AI wrongly detect the Object (for example, the date 14 April, the AI detect it as 1A April.)

This inconsistent behavior make the customer asked us to not use the AI.

Hello garecanbp ,

A few items:

1. You don't have to exclusively use AIUtil or properties-based, you can mix and match, using whichever is better for you at that particular time on that specific object.  For this particular object on which you are having problems, use the AIUtil rather than spending time figuring out the technical reason that the app isn't exposing the object properties.
2. If something isn't being recognized the same the 2nd iteration, it is probably because visually, the screen is different on the subsequent iteration.  Usually, I've found this to be the case where on the 2nd iteration, the object is no longer unique on the screen as you have described it (like the second iteration, there are now 2 clear cart/delete buttons because now there are multiple products in the cart and you have to anchor off of the product text for what you want to delete).
3. In the case you listed about inaccurately detecting the text, this is part of the OCR.  Keep in mind that the question "What do you see on the screen?" and "Do you see "14 April" on the screen?" are two very different questions.  Chances are good that if you looked for "14 April" when the Inspection tool identified it as "1A April", the OCR would still be able to find the object, just with a lower confidence level.  If not, then you could switch the OCR to utilize one of the UFT OCRs available to you, where you could tune the OCR to work for you application (e.g. ABBYY, Tesseract, Baidu).
4. You'll find that once you have tuned/learned how to interact with your application with AIUtil, your scripts are faster to create, need to be maintained much less often (break less), and you can get more automation released, faster.